nvidia-cuda — Claude Code skill: what it does, how to install
This skill gives an agent CUDA, PTX, Blackwell, and CUTLASS references. Use it when working through GPU code or planning CUDA kernels.
npx skills add Fandhe-AI/agent-reference-skills --skill nvidia-cuda -a claude-codeWhat nvidia-cuda does
nvidia-cuda supplies reference material for GPU programming with CUDA C++ and CUDA Python. Its coverage includes kernel development, nvcc, the CUDA memory model, Unified Memory, CUDA Graphs, Cooperative Groups, the Driver API, and multi-GPU systems.
The material also covers PTX ISA, state spaces, MMA instructions, Blackwell tuning, Streaming Multiprocessors, and NVLink. CUTLASS sections include CuTe DSL, C++ GEMM, Tensor Cores, tcgen05, task scheduling, and the Operator API. nsight-compute and compute-sanitizer are included among the tooling references.
This skill is marked as not intended for direct user invocation, so it does not provide a dedicated slash command. Claude Code installation uses npx, with the installed skill file at ~/.claude/skills/nvidia-cuda/SKILL.md.
What it helps with
- inspect a CUDA kernel written in C++ or Python
- locate PTX ISA guidance for a low-level GPU issue
- compare Unified Memory with CUDA Graphs for a design
- plan a GEMM implementation with CUTLASS or CuTe
- find debugging references for GPU code
Example prompts for the agent
- Review this CUDA kernel and point me to the memory-model sections that apply.
- Compare Unified Memory with CUDA Graphs for this design and describe the code changes.
- Find the CUTLASS GEMM references relevant to a project using
tcgen05.
How to install nvidia-cuda
1Skills CLI (Vercel)
npx skills add Fandhe-AI/agent-reference-skills --skill nvidia-cuda -a claude-code2Manually: put the files in place
~/.claude/skills/nvidia-cuda/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/nvidia-cuda/
1Skills CLI (Vercel)
npx skills add Fandhe-AI/agent-reference-skills --skill nvidia-cuda -a codex2Manually: put the files in place
.agents/skills/nvidia-cuda/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Fandhe-AI/agent-reference-skills --skill nvidia-cuda -a cursor2Manually: put the files in place
.cursor/skills/nvidia-cuda/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Fandhe-AI/agent-reference-skills nvidia-cuda2Manually: put the files in place
.github/skills/nvidia-cuda/SKILL.md1Skills CLI (Vercel)
npx skills add Fandhe-AI/agent-reference-skills --skill nvidia-cuda -a gemini-cli2Platform CLI
gemini skills install https://github.com/Fandhe-AI/agent-reference-skills --path skills/nvidia-cudaManually: put the files in place
.devin/skills/nvidia-cuda/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add Fandhe-AI/agent-reference-skills --skill nvidia-cuda -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about nvidia-cuda
How do I install nvidia-cuda in Claude Code?
Run npx skills add Fandhe-AI/agent-reference-skills --skill nvidia-cuda -a claude-code, then check ~/.claude/skills/nvidia-cuda/SKILL.md.
What does the nvidia-cuda skill do?
It gives the agent references for CUDA C++ and CUDA Python, PTX ISA, Blackwell architecture, multi-GPU systems, and CUTLASS and CuTe libraries.
Which platforms support nvidia-cuda?
The skill is intended for Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and OpenCode. It is not set up as a direct slash-command skill.
